Reference · 35 terms

The plain-language AI glossary

Every term in one or two honest sentences. Search below, or skim A to Z.

— A —

Agent

An AI that doesn't just answer, but takes actions for you: browsing the web, using apps, writing and running code, completing multi-step tasks with limited supervision.

AGI (Artificial General Intelligence)

A hypothetical AI that matches or exceeds humans across virtually all intellectual tasks. Today's AI is impressive but not this — when you hear AGI, think 'the debated future,' not 'the current product.'

AI (Artificial Intelligence)

Software that performs tasks we associate with human intelligence: understanding language, recognizing images, reasoning, creating. Today the word usually refers to generative AI.

Alignment

The research field focused on making AI systems behave helpfully, honestly and safely — doing what humans actually intend, not just what was literally asked.

API

A way for programs to talk to other programs. Developers use AI company APIs to build AI features into their own apps; you indirectly use APIs constantly.

— B —

Benchmark

A standardized test used to compare AI models — like exam scores for AIs. Useful but imperfect: a high benchmark score doesn't guarantee the model is best for your task.

— C —

Chain of thought

When a model works through a problem step by step before answering. Reasoning models do this automatically; you can also just ask any model to 'think step by step.'

Chatbot

A conversational interface to an AI model. ChatGPT, Claude and Gemini are chatbots; the models underneath have their own names and version numbers.

Context window

The amount of text a model can 'see' at once — your conversation, pasted documents and its replies. Exceed it and the model starts forgetting the earliest parts.

— D —

Deep learning

Machine learning using large neural networks with many layers. It's the engine behind essentially all modern AI.

Diffusion model

The technique behind most AI image generators: it learns to turn pure noise into a coherent image step by step, guided by your text description.

— E —

Embedding

A way of turning text (or images) into lists of numbers that capture meaning, so software can measure how similar two pieces of content are. Powers AI search and recommendations.

— F —

Fine-tuning

Taking a trained model and training it a little more on specialized examples, to make it better at a particular task, domain or style.

— G —

Generative AI

AI that creates new content — text, images, code, audio, video — rather than just classifying or analyzing existing content.

GPT

'Generative Pre-trained Transformer' — OpenAI's model family name, as in GPT-5. Often used loosely to mean AI chatbots in general.

GPU

The computer chips (graphics processing units) that AI runs on. When you read about compute, data centers or chip export rules, it's about these.

— H —

Hallucination

When an AI confidently states something false — an invented fact, citation or detail. The single most important limitation for beginners to know about.

— I —

Inference

Running a trained model to get answers (i.e., every time you chat). Distinct from training, which is the expensive learning phase that happened beforehand.

— J —

Jailbreak

A prompt designed to trick a model into ignoring its safety rules. Mostly relevant as a news term; model providers continually patch against them.

— K —

Knowledge cutoff

The date when a model's training data ends. Without web search, it knows nothing that happened after this date.

— L —

LLM (Large Language Model)

A neural network trained on vast amounts of text to predict language. The core technology behind ChatGPT, Claude, Gemini and most modern AI tools.

— M —

Machine learning

The broad approach of letting software learn patterns from data instead of following hand-written rules. AI is the goal; machine learning is the method.

MCP (Model Context Protocol)

An open standard that lets AI assistants connect to outside tools and data sources — calendars, databases, company files — in a consistent way.

Multimodal

A model that handles more than text: it can look at images, read PDFs, hear audio, sometimes watch video — and often produce these too.

— N —

Neural network

A mathematical structure loosely inspired by the brain: layers of simple units that, trained together at scale, learn remarkably complex patterns.

— O —

Open weights / open source

Models whose internals are published so anyone can download, run and modify them (e.g. Llama, DeepSeek). The alternative is closed models accessed only through a company's service.

— P —

Parameters

The internal numbers a model adjusts during training — its learned knowledge. Modern frontier models have hundreds of billions to trillions of them.

Prompt

Whatever you type to an AI. The skill of writing good ones — clear context, task, format — is called prompt engineering, and it's learnable in an afternoon.

— R —

RAG (Retrieval-Augmented Generation)

A technique where the AI first looks up relevant documents (your files, a knowledge base, the web) and then answers using what it found. Reduces hallucinations and adds fresh or private knowledge.

Reasoning model

A model that 'thinks' before answering — generating internal step-by-step work on hard problems. Slower but markedly better at math, logic, planning and code.

— S —

System prompt

Hidden instructions given to a model before your conversation starts, defining its role, tone and rules. Custom assistants are mostly just a crafted system prompt.

— T —

Temperature

A setting controlling randomness. Low = consistent and focused; high = varied and creative. Chatbots pick sensible values for you; API users tune it.

Token

The chunks models read and write — roughly ¾ of a word each. Pricing, speed and context limits are all measured in tokens.

Training data

The text, images and code a model learned from. Its strengths, blind spots and biases all trace back to what was (and wasn't) in here.

— V —

Vibe coding

Building software by describing what you want in plain language and letting AI write the code — iterating by feel rather than programming line-by-line.